Skip to main content
Article
Fast Multi-Channel Gibbs-Sampling for Low-Overhead Distributed Resource Allocation in OFDMA Cellular Networks
Department of Electrical and Computer Engineering Technical Reports
  • Po-Kai Huang, Electrical and Computer Engineering, Purdue University
  • Xiaojun Lin, Electrical and Computer Engineering, Purdue University
  • Ness B Shroff, Ohio State University
  • David J. Love, Purdue University
Abstract

A major challenge in OFDMA cellular networks is to efficiently allocate scarce channel resources in order to optimize global system performance. In particular, the allocation problem across cells/base-stations is known to incur extremely high computational and communication complexity. Recently, Gibbs sampling has been used to solve the downlink inter-cell allocation problem with distributed algorithms that incur low computational complexity in each iteration. In a typical Gibbs sampling algorithm, in order to determine whether to transit to a new state, one needs to know in advance the performance value after the transition, even before such transition takes place. For OFDMA networks with many channels, such computation of future performance values leads to a challenging tradeoff between convergence speed and overhead: the algorithm either updates a very small number of channels at an iteration, which leads to slow convergence, or incurs high computation and communication overhead. In this paper, we propose a new multi-channel Gibbs sampling algorithm that resolves this tradeoff. The key idea is to utilize perturbation analysis so that each base-station can accurately predict the future performance values. As a result, the proposed algorithm can quickly update many channels in every iteration without incurring excessive computation and communication overhead. Simulation results show that our algorithm converges quickly and achieves system utility that is close to the existing Gibbs sampling algorithm.

Date of this Version
10-15-2013
Citation Information
Po-Kai Huang, Xiaojun Lin, Ness B Shroff and David J. Love. "Fast Multi-Channel Gibbs-Sampling for Low-Overhead Distributed Resource Allocation in OFDMA Cellular Networks" (2013)
Available at: http://works.bepress.com/po-kai_huang/3/